機器學習方法在股票下市公司之特徵分析應用

在這個科技日新月異產業結構快速改變的時代有一些公司崛起並成為獨占一方新一代的公司，反之，也有些公司會成為時代的洪流，產業並不會停止或倒退，而是會不斷加速前進，無法與時俱進的公司終將會被淘汰。本文資料選自財報狗及 Goodinfo!台灣股市資訊網從2014年12月至2021年7月間共計58家公司11項財務報表分別轉化為特徵分析指標斜率(Slope)、中位數(Median)、平均數(Mean)及標準差(Standard Deviation, Std)後加入石賀元(2020)所提出之 S-score 並利用機器學習方法做為預測公司是否會下市。本研究使用了邏輯斯迴歸(Logistic Regression, LR)、隨機森林(Random Forest)、支持向量機 (Support Vector Machine, SVM)及eXtreme Gradient Boosting(XGBoost)四種機器學習方法來做為公司是否下市之預測，後續也針對不同模型使用了eXtreme Gradient Boosting (XGBoost)內建之函數 Important Feature來篩選出重要變數並加以評估不同模型篩選之重要變數是否能有效的去提升預測公司是否下市模型之準確率。

關鍵字

下市公司；特徵分析； S-score ；機器學習

並列摘要

In this era of rapid technological change and rapid changes in the industrial structure, some companies have risen and become the exclusive new generation of companies. On the contrary, some companies will become the torrent of the times. The industry will not stop or go backwards, but will continue to accelerate forward. Companies that keep up with the times will eventually be eliminated. The data in this article are selected from Statementdog and Goodinfo! Taiwan Stock Market Information Network from December, 2014 to July, 2021, a total of 11 financial statements of 58 companies were converted into characteristic analysis indicators slope, median, mean and standard deviation are added to the S-score proposed by Shi (2020) and machine learning methods are used to predict whether the company will be delisting. Four machine learning methods including Logistic Regression (LR), Random Forest (Random Forest), Support Vector Machine (SVM) and eXtreme Gradient Boosting (XGBoost) are used to predict whether a company will be delisting or not. Thus, this article used the function Important Feature built in eXtreme Gradient Boosting (XGBoost) for different models to screen out important variables and evaluate whether the important variables screened by different models can effectively improve the accuracy of the model for predicting whether the company will be delisting.